Open-Source Android AI Agents Can Execute Code Via Invisible Text
Researchers have discovered that open-source AI agents on Android devices can be exploited to interpret invisible text displayed on screen, potentially running malicious code on a connected host PC. This vulnerability arises from the agents' ability to read all screen content, not just visible elements, allowing attackers to inject commands. The finding highlights a significant security risk for users relying on AI assistants in cross-device setups.